−10.9%
since Sep 30
The idea
I believe AI is moving out of the cloud and onto the devices we already own, as open models shrink enough to run well on local chips. To capture that, I hold the makers of those devices, the edge silicon that powers on-device inference, the memory and storage that hungrier local models require, and a smaller slice of the open-model publishers and their chip fabricator. I'm spreading bets across many names rather than betting on one winner, keeping some cash to add on pullbacks, and treating this as a two-year view rather than something to hold forever.
What it owns
- Devices running AI models locally need to be designed, manufactured, and sold at scale, creating demand for the companies that make consumer PCs, enterprise laptops, compact servers, and consumer GPUs. · AppleApple runs AI inference on its own devices using its neural processors, and the company's large installed base means expanding on-device AI reaches hundreds of millions of machines., NvidiaRTX GPUs are the standard hardware for running open-source AI models at home, so demand for consumer graphics cards grows as local inference becomes more common., Dell TechnologiesDell sells enterprise and consumer laptops that get refreshed when businesses upgrade to AI-capable machines, so corporate AI adoption flows into its PC shipments and pricing., HP Inc.HP is one of the largest PC manufacturers, so an AI-driven refresh cycle in laptops and desktops directly increases its unit shipments and pricing., Super Micro ComputerSuper Micro builds compact on-premises servers for businesses running AI locally instead of renting cloud capacity, capturing demand from edge and on-prem inference., Corsair GamingCorsair sells memory, power supplies, and cases for enthusiast-built AI PCs, representing the consumer and DIY segment of local model inference.33%
- Processors and chips designed to run AI inference on devices rather than in the cloud drive the take, including mobile NPUs, programmable logic, and connectivity chips that edge devices depend on. · QualcommSnapdragon processors with dedicated NPU cores go into Windows laptops and premium phones, making Qualcomm a core supplier for AI-capable mobile and portable devices., Advanced Micro DevicesRyzen AI chips with built-in NPUs compete for share in laptops and desktops, and AMD's consumer GPUs also power home-based model inference., BroadcomBroadcom supplies networking and connectivity silicon that edge devices depend on, plus custom chips for companies building their own AI processors., Arm HoldingsArm-based chip designs dominate phones and a growing share of laptops, so expanding AI on these devices means more royalties for Arm from each processor sold., Lattice SemiconductorLattice makes low-power programmable chips that run AI in cameras, edge devices, and industrial gear where power efficiency and small size matter more than raw compute., AmbarellaAmbarella's chips run computer vision AI inside cameras and vehicles where inference must happen on the device itself by design.27%
- Local AI models require more memory and storage capacity per device than previous generations, so companies supplying DRAM, NAND flash, and storage drives benefit from that hardware refresh. · Micron TechnologyLocal AI models consume more DRAM than previous PC generations, and Micron supplies the memory inside those devices, so higher memory content per machine drives its sales., Western DigitalAI models and data stored locally live on hard drives, and Western Digital supplies the storage devices that must expand as AI workloads grow., Seagate TechnologySeagate supplies drives that store the data local AI systems create and consume, both in homes and at small edge servers near users., SandiskSSDs in AI devices rely on NAND flash memory, and Sandisk supplies the flash chips that go into laptops and phones running local inference.19%
- Open-source AI models running on local hardware create a foundation for on-device inference, and the manufacturers enabling this ecosystem benefit from the resulting device and chip volumes. · Meta PlatformsLlama open-source models make capable AI inference available on local hardware, and Meta's continued release of competitive models sustains the entire on-device AI premise., Taiwan SemiconductorTSMC manufactures nearly every chip in this basket regardless of the chip designer, so on-device AI volume translates directly into foundry volume and revenue.11%
- Cash10%